What is the clinical evidence from this study?

Study design: Observational. Population: Cardiovascular state (n=38686). Intervention: Cross-modal autoencoder framework vs. Unimodal autoencoders or supervised deep learning. Primary outcome: Phenotype prediction and modality translation.

What does this research mean for the field?

A cross-modal autoencoder framework integrating ECG and cardiac MRI data improves the prediction of cardiovascular phenotypes from a single modality and enables the imputation of cardiac MRIs from ECGs. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

April 28, 2023Open Access

Cross-modal autoencoder framework learns holistic representations of cardiovascular state

ARAdityanarayanan RadhakrishnanBroad Institute SFSam FriedmanBroad Institute SKShaan KhurshidElectrophysiology

Key Result

A cross-modal autoencoder framework integrating ECG and cardiac MRI data improved the prediction of cardiovascular phenotypes from a single modality and enabled the imputation of cardiac MRIs from ECGs.

Study Design

Type

Observational (n=38,686)

Structured PICO

Does a cross-modal autoencoder framework improve phenotype prediction and enable modality translation and unsupervised GWAS using paired ECG and cardiac MRI data?

Population

38,686 paired median 12-lead ECGs and 50 frame videos of long-axis cardiac MRIs from the UK Biobank

Intervention

Cross-modal autoencoder framework for integrating distinct data modalities

Comparator

Unimodal autoencoders and supervised deep learning methods

Outcome

Phenotype prediction performance, modality translation accuracy, and identification of genotype-phenotype associationssurrogate

A cross-modal autoencoder framework integrating ECG and cardiac MRI data improves phenotype prediction from single modalities and enables unsupervised discovery of genotype-phenotype associations.

Limitations

UK Biobank samples are limited in their diversity with individuals primarily falling between the ages of 40-69.
The UK Biobank is known to contain racial and socioeconomic biases.
Confounders such as age and sex are embedded in the representations and must be carefully removed for genetic analyses.

Abstract

A fundamental challenge in diagnostics is integrating multiple modalities to develop a joint characterization of physiological state. Using the heart as a model system, we develop a cross-modal autoencoder framework for integrating distinct data modalities and constructing a holistic representation of cardiovascular state. In particular, we use our framework to construct such cross-modal representations from cardiac magnetic resonance images (MRIs), containing structural information, and electrocardiograms (ECGs), containing myoelectric information. We leverage the learned cross-modal representation to (1) improve phenotype prediction from a single, accessible phenotype such as ECGs; (2) enable imputation of hard-to-acquire cardiac MRIs from easy-to-acquire ECGs; and (3) develop a framework for performing genome-wide association studies in an unsupervised manner. Our results systematically integrate distinct diagnostic modalities into a common representation that better characterizes physiologic state.

اسأل الذكاء الاصطناعي

Bookmark

View Full Paper